Sentiment Classification Using Graph Based Word Sense Disambigution

In recent years, with the rapid growth of social media, such as forums, blog, discussion boards and social networks, people can freely express and respond to opinion on variety of topics. Reading and understanding of the huge amount of reviews are not possible for individuals and companies. Opinion mining and sentiment analysis aims to extract, process of the opinionated text and present them friendly to users. Sentiment classification is the most active field in opinion mining that aims to determine whether an opinionated text expresses a positive, negative or neutral opinion. Existing lexicon based sentiment classification methods are unable to deal with context or domain-specific words. To solve this problem, Word Senses Disambiguation (WSD) is useful to identify the most related meaning (sense) of a word in a sentence. In this paper, a sense level sentiment classification method is proposed that determine the sentiment polarity of words using graph based WSD algorithm and a multiple meaning (sense) sentiment lexicon. To evaluate the impact of WSD on sentiment classification, the proposed method compared against a baseline method using two subjectivity lexicons, namely the MPQA and SentiWordNet. Experimental results using a benchmark dataset show that the WSD is effective for sentiment classification. KEYWORD opinion mining, sentiment classification, word sense disambiguation, context dependent word